The Average Scenarios dataset averages all the numerical columns of the scenarios into one row, outputing one row for each Location, Year, and RCP. This dataset is used when conducting EDA and visualizing overtrend
The data is collected over 113 locations within the Natural Bridge National Monument. (Number of Unique latitude, longitude combinations)
This dataset is composed of 41 years of historical data and 79 years worth of predictions. Since there can be only one scenario for past data, all historical data is labeled as ‘sc1’ or scenario one
From the predicted years (2021 to 2099), There are two RCP scenarios which changes only the RCP variable and 40 scenarios which simulate 86 other variables.
Based on each combination of scenarios, a prediction is made for each location point regarding various columns of the dataset including annual and seasonal percipitation, seasonal dry soil days, seasonal evaporation, seasonal extreme short term dry stress, soil water availability to output a final prediction for Annual and seasonal temperatures.
What is RCP?
Representative Concentration Pathways : A group of scenarios where CO2 emmission is predicted like the image below
The dataset consists of two RCP scenarios 4.5 and 8.5
source : Representative Concentration Pathway. (2024, May 2). In Wikipedia. https://en.wikipedia.org/wiki/Representative_Concentration_Pathway
Location
Where is this data located and how does it look like?
The data points were sampled at the Natural Bridge National Monument in Utah. And for a better idea of how temperatures and vegetations are distributed, the plots below show two different location aspects of the dataset. The first plot is the average annual temperature for each location point in the year 2099. Since the temperature for predictions increase over time, the last year for the dataset was chosen for a more dramatic comparison
The second plot is a scatter plot of the locations of vegetations. By comparing the two graphs, we can tell that there isn’t much correlation with vegetation and annual temperature but rather a correlation with the location(latitude, longitude) and temperature. We will get to this in the following visualizations.
The following plots were drawn by averaging all scenarios, locations, and RCPs for a given year for annual temperature and annual percipitation to see the overall trend of the predictions of the dataset. Predictions were made from the year 2021 which is
We can see that the predictions portray an increase in temperature but a fluctuation with percipitation allowing us to make an educated guess that temperature is the more important variable for RCP scenarios which deal with CO2 emission.
What are some aspects of the datasets we can slice and dice or drill down to compare and retrieve meaningful insights?
Most numerical features are generated based on the scenario of the model that generated future data, and some numerical features such ase S,E,Slope, RF, Cl, Sa, El, treecanopy etc. are features that are fixed according to a unique location. Therefore categorical variables are the aspects of the datasets we can compare to create new insights
Categorical Variables :
RCP
Vegetation
Scenario
The following plots compare the predicted annual temperature for each category for the three categorical variables
Since RCP deals with CO2 emission and the 8.5 scenario has a higher emission prediction than the 4.5 scenario, the annual temperature increase of RCP 8.5 is more rapid than rcp4.5 although both are increasing.
Temperature comparison (Vegetation)
# Filter the DataFrame for 'RCP' values 'historical' and 4.5filtered_data = df_con[df_con['RCP'].isin(['historical', 4.5])].groupby(['year','veg'])['T_Annual'].mean().reset_index()fig = px.line( data_frame=filtered_data, x='year', y='T_Annual', color='veg', title='<b>Annual Temperature by Vegetation Type</b>', labels={'T_Annual': 'Annual Temperature'})fig.update_layout(title={'x':0.5})# Add a vertical line at year 2021fig.add_shape(dict(type='line', x0=2021, y0=filtered_data['T_Annual'].min()/1.1, x1=2021, y1=filtered_data['T_Annual'].max()*1.1, line=dict( color="Red", width=2, dash="dash", ), ))fig.add_annotation(dict( x=2021, y=filtered_data['T_Annual'].max(), xref="x", yref="y", text="Prediction", showarrow=False, font=dict( size=12, color="Red" ), align="center", xanchor="left" ))fig.show()
The vegetations seem to follow exactly the same pattern of prediciton with a fixed difference between each other. This may mean that when calculating predictions based on scenarios, the algorithm was modeled so that the mean of the vegetations were always a given distance apart from each other. Because of this limitation of the algorithm, it is unncessary to compare vegetations from each other. We will always get the same difference.
Since we already know that RCP plays a big role in how the algorithm predicts the temperature, We will group the dataset into RCP4.5 scenarios and RCP8.5 scenarios to see if there is a significant difference. By examining the plot we now know that RCP 4.5 correspons to scenario 22~41 and RCP 8.5 correspons to scenario 42~61. There are cases where 4.5 scenarios had higher temperatures than 8.5 scenarios, but since RCP acts as the first drill down layer of the dataset, we can use the scenario column as the second drilldown of the dataset.
Statistical Significance
Is there a significant difference between different scenarios?
Before we start analyzing our dataset, one final step we want to take is proving the statistical significance in the different scenarios we plan on comparing.
The three comparisons we plan on making are as follows: 1. RCP 8.5(High) vs RCP 4.5(Low) 2. RCP 4.5 : Scenario 37(High) vs Scenario 40(Low) 2. RCP 8.5 : Scenario 60(High) vs Scenario 58(Low)
For our dataset analysis, we will be comparing the maximum and minimum scenario for each RCP group to analyze what features affect temperature the most. That is comparing scenario 37 to scenario 40 for RCP 4.5 scenarios, and comparing scenario 58 to scenario 60 to do the same for RCP 8.5.
Next Steps!
Now that we’ve proved that the difference between RCP scenarios, and the highest and lowest scenario for each RCP group are all statistically significant, lets dive deeper into the dataset to construct visualizations to hypothesize features that have correlations to the predicted temperature!
